very much for the introduction
uh
we talk about anomaly detection
which is a topic which is being around one time
uh the reason why i'm interested in this topic is that the
so we have a national
project
the major object
which it is addressing the issues both you based on the computer vision system
other
the main application slight changes
what do you do you have to start from scratch
all you have to can you use some of the models and uh one of
the issues that one
in this context
it's uh on the detection because
the system has no
that if it is fully automatic system
because you know that the it cannot cope with
the main in both uh that uh because no competence to in that the
since the data so that's the context and because it's a reasonably project
we in the groove in psychology of the community college london
so the plan is the
stop
the background then we want to on the money detection
uh
we review all uh right out on anybody detection and that
a little bit of it is
all
approaches
and that will then be all position
yeah
solely on the money detection
section system channel
and the
we apply
oh set
the problem
you know
interpretation system
so that's plan to
so if you
this on vision system we present system the difficult to a stage is and the
first of all the to the remote modules
solving lost six
do about the if you are not just like to do not basically problem but
uh
image processing vision i want to see you developing a system that actually application and
many other issue
think about the channel
you need to collect a lot of training data because the existing systems uh i
let me
observations
we do not know what is
indicating that
and the optimized system so it's like that
uh nobles that's the goal go through an image that is why convolving
and just uh as an example are we talking about the tennis video analysis
you
for some
school
yeah so uh and that was just a very few men version of this is
that the linear
and uh so it's at a G is to you
an application and then
all services about i
okay so um
the conference here is the uh is concerned with advanced concepts and in a way
when you develop uh and interpretation system then uh in the sense that system is
advanced in its own right so i could be just talking about the video uh
the tennis video notation system but then my focus will be more on the second
body point
as i already mentioned so suppose you want to add up the system to some
other domain uh even quite close domain and go see that the applications i will
be uh talking about a very simple indeed nevertheless uh raising like you interest in
issues and challenges
and that if you want to go that if you want to
benefit from many years of after and then try to use what you have and
to develop a new uh competence you capability than possible you have to
identify that you have a problem that you cannot cope with some input and uh
then you have to modify the system inappropriate way and there are of course the
other communities at all this stuff community support computer vision that whether or not a
transfer that i mean and uh so uh
will not be addressing those issues but uh at the end that once you have
adopted the system and some new application then
when i say i'd update
i really mean develop new capability then the system needs yeah and not the functionality
it needs to know uh can make sure a situation it is operating and that
should be able to classify the context and uh in which it operates so that
it can automatically select the appropriate uh domain knowledge voice separation so
this is the system that we developed so basically it's the can analyze tennis video
the way uh
that we describe what the system looks like by the in principle
the objective is that uh from the video it input completely automatically you are able
to interpret what's going on to the point of points awarded avoiding the uh generating
school from the process now
i'm not talking about the uh style whole uh yeah we develop a system which
works that from two D standard the real cost video okay so that makes a
problem it would be difficult but anyway so in principle when you break the video
into shots you want to know what's happening in short
well as or so seconds uh and that there is not only uh who actually
means in the running and we should be awarded a point
no
probably unless you are young and have very good a nice and uh you will
not be able to see the detail about the this is just to illustrate the
complexity of the system
and that it has uh why the few levels of course in so initially the
uh video is broken into shorts and then the short each shot this process the
separately basically uh and that is the
level processing deals with the foreground-background separation
then the key components of the content are extracted which is the motion of the
ball and the players and the then the system yeah that uh means uh important
events
and which is uh one important event is when the board changes detection and way
it changes direction
and then eventually there is some high level interpretation process of these talents so this
is a more digestible somebody of the system okay about that basically the ball tracking
is the most important you need to know whether code is uh you need to
the text is important events and there is a high level interpretation part which is
basically hidden markov model based
no most of the modules that the system has use context in some way okay
so when i talk about context here it's not the context it's not the domain
where the system operate but it's the local context which is like the temporal or
spatial so when you want to interpret for instance uh what's going on need to
know not only whether board is but also whether players are so uh that is
the interaction between objects in the video uh so in principle you are interested in
integrity in every object in each frame about the neighboring objects have a uh
one may also information which is which is very important and you want to use
this to uh information jointly uh they provide contextual information and you want to use
this information jointly to make interpretation so in principle you have some slow but knowledge
domain knowledge which is a quite in some way i the through line in or
partly through
yeah so you didn't in the prior knowledge in uh and you are then comparing
observations ritual model to make interpretation so this is very genetic uh indication that most
of the modules are dealing with contextual information many more usability contextual information uh over
time okay so that uh about the other modules deal with the spatial contextual information
and some of them with both
so the first one for instance is a module which is uh separating foreground and
from background so you may want to what happened here uh
because players disappear but basically it's the module which is below the remote site so
you take video frames from a shot and the and relate them to each other
and uh basically that allows you to go to was i and anything that's movie
that frame is wiped out because it uh not the assistant information and so you
have basically a background and then you can use the background to separate the foreground
probably so
that's one example all the that all this type of functionalities that the modules perform
the most important one once you have uh
uh the players and the can extractable used to detect the events so you can
see that the so it's the ball tracking problem uh process and that each is
also detecting when the ball is changing detection and uh you know uh where the
code is that has been automatically the big picture it's a fully automatic system we
can uh and that you also can detect players and from that you can derive
interpretation
this is uh
so these are the events that we have extracted in time and the
the sequence of these events and the position but they happen any action or more
advanced a bit plane is a determine what's going on and you have a hidden
markov model but it's a lot of the temporal structure in a small gains in
general and so the mean pennies uh which allows you to interpret what's going on
and you can then decide to who should be awarded to point at the end
okay so and this is an example of what the system would produce so he
on the left hand side you to actually tell you what's going on was awarded
the point at one time at a tool training
okay so we as i said you spent three years developing the system and we
were just working with one video and it happened to be a video singles
and then a somebody else question about what would happen if you actually applied it
to doubles and you know so it's very simple the small transition but the nevertheless
at uh
significant enough transition for the system to fail so uh and uh
so that's one thing about the
it's not only question all system fail in you also would like to know uh
when it fails to white fellows and can use land or something from it
anyway so the question is what are the mechanisms that are needed for the system
i didn't to realise that it's actually no longer competent to perform a certain functionality
and the how can this functionality be extended
already mentioned so this is the project the that we have features been sort of
a motivating the work in this area and the anyway so already i think alluded
to these mechanisms that we need to i don't to take this we need to
cross knowledge and the we need the to adapt interpretation processes and acquire new competencies
that way
okay so
these are the mechanism this is done is and to what i'm going to focus
on anomaly detection so already talked with twenty minutes and i haven't the restarting the
topic of the of the lecture okay so uh these are the mechanism that would
be normally needed and that but one of the nice anomaly detection
oh if you look at
the
it well as the definition of on the money to start with and it's a
normally understood this um so something deviating from automatically but the that the how the
normal it is defined yeah is very general and that can be some sort order
it can be sort of a statistical normally you can be a rule whatever so
it's uh original there are also many synonyms and the interestingly some of these uh
pseudonames the general mean
deviation from normality about the sometimes the uh they have some uh additional nuance uh
and that they may need for in cincinnati
yeah regularity okay innovation so there is a
difference between uh and the money and innovation because innovation usually means implies a change
is of constant change you moving to some of the uh model of a proxy
experience
now what is that conventional model i think everybody knows that the menu look what
anomalies you are normally thinking in terms of uh outliers of some distribution uh so
you have a gaussian for instance and that was the
uh making observations away yeah then used several it must be applied must be anomalous
observations because it's not pretty consistent with my model of the data the experience the
time uh that i make the past so one is a
look in uh and basically the mathematical model is a statistical one in principle and
the uh
sometimes you the not only work with a single observation but the weight the multiple
observations and then you may be interested whether uh we distribution of the all observations
are different from the distributions of but uh of your model and uh so you
could also be talking about the sum so that uh normally in terms of the
shape of the distribution
as i said to anomaly detection has been of interest for a long time uh
domain and value goes back to the nineteenth century a people have been interested in
developing normal model so gaussian models and uh for model in various uh sets of
data observations and the and how they have been detected by the model is uh
when the observation is consistent with that model so over the uh hundred years i
suppose most of the work has been focusing on this type of concept of but
uh no money and there are excellent surveys which uh make like quite easy and
uh recently quite a lot of working in on the money detection comes from the
security and the surveillance the communities as they are very much interested in formulating the
problem of but uh detecting the something unusual as the and on the water detection
problem but that although they may be using quite complex system most of the uh
notions of on the money in these the papers are very close to the statistical
notion so even if you have a complex just images multiple layers of interpretation very
often people still uh loop on the money from these the uh from these models
so you can estimate are presented in a very simple way is here so this
is your basic system which is performing sometimes you have sense uh you got some
usually single hypothesis model
uh so i could distribution and the there and uh this derive some action something
that something and you are interested to know whether the uh that is any and
all money so you need some sort of a anomaly detector and usually would be
some sort out lie detector and if it is an outlier then hopefully it will
affect the action so you will not but for what you would normally performed
no in a complex systems like uh a video system tennis video system you need
to model like this big every model okay
many of these modules are dealing with the multiclass problems so you don't have just
a single
hypothesis you have multiple hypothesis which is also introduced in the interest in complexity the
into the equation you have a
many levels of course in and some of these models are delay in a weighted
high level information they have that down uh using contextual information and uh so although
they may be interpreted the same sort of a have and they will be using
different sources of information and so all these uh complexities are somehow not cultivate indicated
weighted by these dimensional anomaly detection uh model so already mentioned so this the list
of things so we have multiple models not just a single white with two hypotheses
model
importantly in a much in perception
very often we use discriminative approaches rather than generically if using discriminative approach you cannot
really talk about outliers because you just know whether things on the right side of
the boundary on all but the you have completely lose the uh every idea of
that the observation which the which are trying to classify as an outlier on all
is lost the uh to the system so um and if you wanted to detect
a normally
you would need to use both discriminative models get better performance but also maintain a
generative model to know what's going on whether you are actually competent to make that
decision
uh you have very often areas in the observation space where you have a genuine
ambiguity now give a genuine on but then the decisions you make you make in
uh you have to be very careful about the menu can not necessarily interpret them
as kind of money because you are you have a ambiguous situation you cannot have
confidence that it's going to be an anomalous observation
contextual reasoning already mentioned that the uh
existing systems are not ready yet to deal with that and hierarchical representation
about the two more things uh data quality you need to know whether the observation
data you wanted and weighted is of the same quality as the data with the
page the system has been designed you know that you make certain assumptions about the
quality of the data any that quality changes then
you the system has to decide if you differentiate between that situation and uh because
it would be starting making errors okay and the anomalous situation where you if you
have good quality data can be pretty confident that if something is the image then
that it's going to be anonymous so the observation
and uh
more the boolean because it's a very often one
introduced is uh
a potential one another situation
by uh
you'll interpretation process because you want make that process to be as fast as possible
so for instance if i am interested in object recognition and i know there is
uh i don't know half a million objects
right at hundred thousand objects you look at the various names and dictionary whatever it
would be completely foolish to have a system which can interpret and very single object
from that hundred thousand one place so you would the room that leads to something
manageable and hopefully we'll deal we just uh i don't have it and the hypothesis
on the list and all than a hundred thousand and that if you do that
then you may observe something which is an autonomous but by your decision because you
have actually simply by the system goes uh processing strategy is and making the assumption
that the object will come only from this subset you yeah and if it doesn't
then you should be able to detect it and recognise it and to do something
about so you can then inject more hypotheses into the system uh if the none
of the existing hypotheses is uh to get
so
i talked about the deficiencies of or not normal anomaly concepts and just to show
you more examples of the different nature all but not on the model situation so
very often
one is ask uh to solve the problem of spotting the difference okay so you
can consider it also as a on the money detection problem so in this particular
situation we have a nice a nice little object and that i think everybody cans
for the difference is a head of a cat hopefully or something uh in the
second picture are there any other animals
very good yeah
uh so this object has slightly different like uh angle any other
yeah and the little bit shifted very good so we are very good on the
money detectors
but the uh the first instance was not all that will be is that all
these uh the other animal is represent about the you know very simple uh comparison
uh and four that's a computer systems are extremely good uh able to detect uh
the dependencies and the you can uh in well okay so that's uh that's one
example you have we already talked about distribution drape you talked about mobile the innovations
anyway what about the this case
are there any other monies
well actually there are no differences the only difference is for maybe actually what to
observe an image of a very acute vision uh what you jobs uh is the
difference in uh information about the second image has been compressed data okay so you
lose a little bit of a high frequency information but uh so obviously the compression
introduces an obvious and if i have a on the money system which is to
detect independence is that based on the sums of assume distribution and uh suddenly the
noise characteristic change then uh you know is that difference not so this should not
be detected as a normal is so big that quality is an extremely important concept
in the in the process
already talked about the
uh contextual information and the or and hierarchical representation speech also exploit contextual information and
uh so you know here
every object in this image which is famous painting uh
make sense is able to find about the relationship of these objects is the obviously
unusual because you would not expect the locomotive to be jumping out of the fireplace
and the uh so
uh it's another example of the type of anomaly that you would like to be
able to detect and
explored and the system should be exploited so this is the conventional system that uh
people have been using them almost four hundred years and um
and this is probably what we need okay so
the difference between that well this is the actual functioning system which is uh implement
in some applications uh this just uh is the same thing is the blue box
which has sensor and the actions alignment
when ten okay
the difference between this and that is that we have a probably multiple hypotheses of
hypotheses the for each uh module okay and the or so we have probably several
layers of interpretation not just a single layer we sure uh
yeah so the high less would be using context and uh so that is the
relationship between those players uh so you then need if you want to the text
on the money in a sensible way you then need the following you need something
that deals with the differences between contextual or non contextual processing
and that that's a soap incongruence detector okay so uh yeah which is so if
you have an object if i go uh back to my
good really uh if i go here
if i and this is my scene graphs or something estimation and in principle i'm
uh trying to interpret every object okay but we know that i am interpreting one
object uh in the to get off then i'm used in the contextual information provided
by other objects so in principle you can uh you are interpreting that object in
two different ways possible just using the measurement information relating to that object
and secondly you use the measurement information and possibly prior knowledge about the configuration of
one or contextual information provided by the neighbours which are will have impact on the
interpretation of the subject so we have soft contextual and non contextual
in the presentation and you can be measured in then continuance between those two
uh
but we need to other things
we need to assess battle or do not actual one and for the contextual one
uh whether we have any but we are dealing with ambiguity so what how much
confidence we actually have in the interpretation that we are making so that's a one
of the things that the needs to be i did in addition to incongruent uh
we need to a module which is a seen data for the because that module
tells us whether we really should be
looking for a normally sober that even if you'd the text something spurious uh whether
we should consider it as a normally because if the data quality has changed then
we should not be uh
simply saying well it's anomalous situation because so uh yeah the
incorrect decisions so what about the change that will be induced by uh data of
different quality uh well we should be you know and the
and in addition to all that we need to the east and that
uh anomaly detection process is the outlier detection process is because even if my non
contextual and contextual decision making process is a uh
functioning well and uh to function well they would be probably based on the stigma
not body models then i will need
some way of method deciding whether the observations a on the models are not whether
they are outliers so i still need to the conventional model okay of undermining so
that can see that these two blocks are the cable uh non contextual and contextual
process
but hopefully i will not be using them very often because if i did lana
the system would just the be computationally complex so uh
ideally what uh you would like to do is to
bros processing in these modules looking for our model is only when you want to
get to do so and this the to get in can be done quite efficiently
why this incongruence detection process
now
can see that one of the mechanisms and only one there are others uh in
uh
the system that we need for detecting a normally scene perception systems is uh incongruence
detect that and interestingly uh the work which uh well one of the original work
in this area uh was running speech area uh
i don't know whether actually brno was involved in this or more uh was it
was just one of yours
okay yeah so you work with the hynek hermansky and um work on the problem
all the out-of-vocabulary what detection which is exactly the sort of a big a typical
example of the problem we are dealing with you may have a uh you have
a at least player speed a system which is processing data uh detecting phonemes so
we have non contextual interpretation and contextual which combines the phonemes in words and you
may be interested in detecting and whether there is any anomaly and that would be
an or more like if for instance the phoneme detector functions that very well gives
you very strong confidence in the interpretation but uh the
word-level interpretation of police is garbage and it reduces got it's simply because the word
doesn't exist in the dictionary
so this is the no example of the situation uh that uh we would like
to detect and the there was a five year project direct project funded by the
U which is uh as being extending this basic idea to the image domain
and the and also continued with application in speech and uh so that was uh
but also by will get which it was then uh extending this work uh and
the most of the other work which are the definitely want role is uh
this name it yet the publications was published in the subsequent about two thousand and
i two thousand and well so
this is a little bit on the background about as i say is not directly
focus and finally on the incongruent so detection how do you uh the fact that
there is a difference between sort of a generic and the specific classifiers generally be
in uh non contextual one uh well depends on the application about the
and the if uh what is the implication of uh detecting such incongruence so that's
uh what dialogue has produced but maybe actually try to use this in a only
work on the tennis video interpretation it was not you know what the very citizen
fine mention be a very open dealing with situations where the decisions but ambiguous and
then you would not a bit on from that come from that you want but
and with a normal situation we dealt with situations and we'll see that in a
minute that the uh we had several videos of pennies and the
even several videos of any single they all had a different chord to the from
different tournaments so uh they had the uh the recorded in different conditions and uh
some of them but noisier than others and that it was pretty a that you
need to know something about data quality if you want uh to make a sensible
uh decisions about on the money we still need it the basically the original uh
technology so to speak of a normally detection so how by detection proces and uh
so i think these were or right and what do they monitoring also is needed
to measure whether distributions of shifted
no wit is uh
architectural system that is the state it it's a quite interesting because you can then
based on the various uh
uh
on the outcomes or on the analysis of the uh the various modules in that
anomaly detection system you can then a classifier you anomalies or situations yeah and they
recognise different states so we can definitely recognise the state when you have no anomaly
but you can also uh identify situations when you are dealing with an unknown up
with noisy measurements you can uh the text situation that you have unknown objects uh
when you have an incongruent or congruent labeling so all the various a space of
uh nobody can be detected and to you get much better idea of what's going
on
so ideally actually what we want to do is to start with ten days and
move on to badminton and uh do uh detector or identify with the modules that
will not have competence to well on the input data and uh try to correct
the module so i don't then all inject knowledge so that the we can actually
use the system volume application
but the
the wise you started something very simple and as i said just switching from singles
tennis doubles so very simple situation so if you consider that problem then
what would you expect
first of all
in doubled there are twice as many players
that's yeah but the cold that is being used for the game is a wider
so you have also the time lines which can uh
which are illegal basically in the case of singles about in the case of doubles
of uh they are more and the but everything else stays the same the rooms
are the same that was that was quite a nice the
uh
challenge because it was not too complicated about the at the same time why the
interesting to see what's going on and uh okay now in principle you would say
well it's obvious well can just count the players and the drop is done about
the impact is anybody who works and you or working in on images or video
you know that the tech T and count been objects it's not as simple as
that uh well lee because
the vision process is are not perfect but partly because the uh application domain allows
basically
uh
well this is not the use of a black and white so we speak about
the it's not either two or four in the game but the there are other
moving objects so you have line charges for instance and normally this tells us they
still and when you uh do the most i can then use of uh they
stay in the image about that sometimes they move okay and if they move they
suddenly become moving object and uh then unless you have some sophisticated mechanism of distinguishing
between players and other moving objects then you are stuck with the different count then
you have more balls okay so the se is played and it goes out and
the more boy runs collectible and uh so you have somebody five
object detected that so if you actually look at and the statistics of a video
okay uh not just the then uh this is what you would to the observed
for singles okay so most of the time you would the detect just to plan
to agents movie nations about the we in the many occasions uh you detect a
human on and uh sometimes up to five so we have a distribution and equally
for doubles uh you have a distribution so you have two sets of this the
uh
you look on the money on the basis of distributions rather than single observations but
anyway so we are basically trying to differentiate between uh
two distributions one which is a modal distribution and one which is of the distribution
and look for differences and that anyway so that's uh what we have a downer
which is a source standard approach and here we have some uh
not the results but the data that we use so we have can see we
have five videos uh of different length so they are not necessary or complete much
is about the white it doesn't that they all of a different situation so we
have uh australian uh japan tournament and us women and men single doubles and these
are the numbers of the place and um
and here we have some results okay so what we show here body to you
as we are comparing distributions if you are using an into information just from one
short then this will give you the performance that you would get
for various scenarios okay and the uh basically uh here we are talking about the
detection of under forty so uh we train on singles and when i talk about
a normally i'm or there's S you mean that any training that is done is
or was down in the norm a normal situation there are many cases where people
are actually trying to synthetic pretty uh genetic on the monies create animal is and
the uh but i think it's fundamentally wrong approach because that if you uh design
a system you cannot possibly collect data or a normal situation for the idea uh
and well then they would just becomes of new classes and the so the really
this they're the appropriate the way of thinking about it is that you cannot train
the system only with the norm on the most data and so order training was
done only on singles we measured the level of noise and you can see for
instance that the was thirty and uh men single pay that much lower high noise
then uh the other two and uh and that uh
uh
has a serious implication because if you look at the data
you can see that the if you train or no uh so here we have
information okay here we trained on the uh australian women singles and japan single okay
so you can see that the
if you train on the uh good quality data and then you try to uh
that's the system with the data of different quality then you have problems in you
can see that from this guitar because this is basically the unwanted detection output or
the single was so we should not be detecting any animal is because the art
doesn't dealing with the same domain the system was trained to but uh to recognise
the right interpret the tennis singles and here we are actually having a problem because
the course of the noise condition uh we are uh detecting force anomalies uh right
is that when we actually use the trained on data which is a little bit
more noisy than that
not all the best uh singles throws any animal is about the uh then we
have to do a little bit more integration to get actually the results uh the
unwanted direction di can correctly so that also shows you that the uh
one is to be very careful about data quality and you just implications on the
on the money detection process
uh the second to the task was to well the second on the money that
can analyze is that the ball goes out in the time lines and
okay and the U
so the gain should terminate
but it doesn't just got it on and uh
again we have developed a so what do we have well we use
uh had be very careful to make sure that the a normal role in us
on the models out who uh situations where uh which may genuinely ambiguous and because
of the data in on the system itself anything very close to the boundary line
between the timeline and the single school was on the models but the further away
you got from that the remote the from the boundary line you have more confidence
so we have values into this a confidence measure
we as a filter to make sure that we are not trying to make uh
decisions about on the money uh on data which is by its very nature i
don't the ambiguous
coming back again to my point that we are always using only the information that
you acquire obtain in the local a problem but uh normal source norm and the
model souls and uh so basically
and the interpretation and the interpretation process associated with it so we have not really
designed the system simply to detect the specifically on the money sits do in normal
processing and the uh detecting on the monies as a result of that and the
anyway
this is a just um an illustration of uh of the interpretation process in the
so when there is a perceived
uh there are okay well as when the system should that i mean a and
actually the game continues we are uh follow in all the possible interpretations uh all
the possible a interpretation possible it may happen and the uh on the basis of
a that we are able to make a decision whether uh there is a no
money because the game continues uh without uh bases and the two
the detection is based on measuring incongruence between
uh contextual a non contextual uh playgirl's basically so we have our event detection which
is uh give you know so non contextual labels and we have the context of
course in which takes into account the sequences uh of events over time so as
uh
as this are normally the case you have basically as i already explained you have
two interpretations one which is contextual non contextual and you have to measure whether they
are incongruent
and one possible way of measuring it is using solve a bayesian surprise measure which
is the form of a divergence on a discrete distributions of labels about the problem
with that the measure is that uh it's very sensitive if you have a uh
a probability which moves from point ninety five one then a suddenly you move into
infinity and it the course this the hubble and uh so we have actually adapted
that mention and to use the something which was a practically a much more efficient
so we chose the top label the most uh the best supporting label for each
of the contextual or non contextual hypotheses and just measure the difference between those two
and you can actually show that in the two class case that we consider in
this particular uh application whether the ball was out not uh we uh it ended
up with a very simple way of measuring an incongruence between the states and when
we did that on the videos that we trained with also we trained on single
us on a single as we had no anomalies detected so no problem as you
would expect and then on doubles uh well with the current system whatever limitations it
has to be certainly detected some anomalies
many where undetected uh not many but is more number of false positives and then
you associate the anomalies with the and you have a cold where they happen
they identified that reminds so it was very nice and that was very easy then
use that association and we have another paper elsewhere uh which uh
and then takes the output of this uh of this module of this anomaly detection
module and through this association is able to but what define the rule based basically
say well the court remove the animal is the cold size has to change and
it has to use that reminds us to uh to be able to in that
discontent successfully so you know eight
i think i
talked about i'll give you examples of all the mechanisms that the rainy day for
anomaly detection and but exercise by application uh principle you need this context detection which
is about domain detection a rather than a uh real or complex uh for system
to acquire new competence and once it has then it has to be able to
pick out which uh domain it's to do it but in a bit and that
the take the appropriate knowledge base and uh this is the basic system is used
in the interpretation that way role of a high level and the this is the
anomaly detection mechanism but that's the module that uh S is still need it and
that would be added to the system to
lexus successfully so that brings me to conclusion i hope that i have a display
did you that uh i know what detection in machine perception requires more mechanism then
what is normally what is just over the body conventional model and the and what
these mechanisms are and how useful in practical applications thank you very much attention
yeah
the use a system
well i think the you know what goes into the anomaly detection system i think
it's genetic about the application was specific okay so obviously are solutions will not work
for your problem about the i think of one uh the notion of data quality
is very important and the also the approach the problem that one needs one should
be trying to train the system just with the normandy time but it's you all
you also mulch within yourself in the foot because if you have examples of on
the money then it would help you to improve the design of the nevertheless uh
you know system then it will be able just to detect the what you presented
to it you in training and uh and so there is a little bit of
a dynamo yeah
i
okay uh
basically i think in all the protocol that all the videos that the use of
uh from professional matches and the cameras with fixed but any okay this is why
we needed to do the most like uh detection with section um
in principle
at least we always use the prior information that this the ground plane so you
need to based on the information you can solve a calibrate the comment on expect
to the scene of the speech and uh so uh you doesn't have to it
can move in it is not the solution is not just for a single position
of the common uh you can always uh contrary the system for any position and
this is what actually happens when a remote uses them
yeah
i think it was more to do with access uh i think the video speech
we go around the through internet ordering to internet maybe unique go but we didn't
looking into it uh but we knew that uh it would be difficult to get
the copies of the same but on broadcast
although we have a one of two with that B C so
a game and uh yeah
it that uh it's not regulate and i think that would say that we have
maybe they are losing probably for the confidence measure we are probably losing half of
the ten timeline
uh the way we are not making decisions because uh
the ambiguity and i'm because we can accuracy of the system and it actually gets
less is for the part of the core okay because the further away from the
comment often a degraded in accuracy
information
well i'm i hope it will generate some other one is but uh that's an
interesting proposition
thus
for
which
and